Microsoft-Backed Space and Time Lets AI Agents Verify Private Data
Microsoft-Backed Space and Time Launches Platform to Let AI Agents Verify Private Data
Not financial advice. Past performance is not indicative of future results. Trading involves substantial risk of loss. Do your own research before making any investment decisions. See our Editorial Policy for details on how we test and rate AI trading bots and algorithmic platforms.
Strip the blockchain branding off this week's announcement and what actually shipped is plumbing for AI trading bots. Space and Time, the Microsoft-backed data warehouse, launched Space and Time Enterprise, a platform that uses its Proof of SQL engine to let AI agents prove that a calculation ran correctly on private business data without ever exposing the underlying records (Decrypt). On its face that has nothing to do with your brokerage account. We think it has more to do with it than the enterprise buyers this press release was written for.
For context on where we sit: our 2020-2026 testing program has run six-month live trials with funded accounts on more than 50 trading platforms and AI trading bots, and we have benchmarked the adaptive engine behind Zephyr AI against that same cohort. Across that sample, the single most common failure we log is not a broken signal. It is an unverifiable input. A bot trades on a data series nobody can audit, produces a P&L nobody can reproduce, and the subscriber has no way to tell whether the strategy that ran was the strategy that was sold. Space and Time is selling a partial answer to that problem, and it is worth understanding exactly which part it answers and which part it leaves wide open.
What Space and Time actually launched
The product is Space and Time Enterprise. The mechanism is Proof of SQL, a cryptographic method for proving that a computation was executed correctly over a dataset without revealing the rows themselves. The company's own summary of the launch is precise: it "uses Proof of SQL to prove calculations on private business data without exposing the records to AI agents" (Decrypt RSS summary, 2026).
Translated into trading terms, that means an agent can be handed a result, say, the average funding rate paid across a set of venues over some window, along with a proof that the number was computed honestly over the data the counterparty actually holds. The agent never sees the raw feed. The counterparty never exposes its book. Both sides get something they did not have before.
This is the closest thing the AI trading bot sector has seen to an audit trail that does not require either party to hand over its crown jewels. Whether it becomes standard infrastructure or stays an enterprise curiosity depends on adoption, and adoption is not something we can measure from a launch announcement.
Why should AI trading bot users care about Proof of SQL?
Because the backtest-to-live gap is, in our experience, mostly a data problem dressed up as a model problem.
When a bot's live equity curve diverges from its backtest, the industry default explanation is "market regime change." That is sometimes true. More often, the backtest was run on a data series that does not match what the live bot actually received: different bar construction, different timestamp convention, different venue, survivorship bias in the symbol universe, or a vendor's adjusted prices that were revised after the fact. None of that is fraud. It is just unverified.
Open-source frameworks make this concrete. Backtrader and NautilusTrader are capable, widely used backtesting and execution libraries, but both rely on the data feed the user supplies. For a quant who controls their own pipeline, that is a feature rather than a bug; the trade-off is that no provenance layer sits between the vendor's CSV and the strategy's decisions unless the user builds one. A proof layer like Proof of SQL is an attempt to make that provenance portable and checkable by a third party.
We cannot publish a backtest-to-live decay figure for any specific bot in this article, because the launch materials do not include one and we will not invent one. What we can say is that in our funded-account trials, the questions we ask first are always the same: where did the historical bars come from, who can reproduce them, and does the provider publish a hash or export that lets a third party re-run the same test.
What we could verify about the launch
The honest answer is: less than the headline implies. Here is the full extent of what the source material supports.
| Claim | Primary source | What we could confirm |
|---|---|---|
| Space and Time Enterprise launched a platform for AI agents to verify private data | Decrypt | Confirmed in the source headline and body |
| Uses Proof of SQL to prove calculations on private business data without exposing records to AI agents | Decrypt RSS summary, 2026 | Confirmed in the source summary |
| Microsoft-backed | Decrypt headline | Confirmed in the source headline |
| Vendor regulatory status in the UK | FCA register search | No matching retail-facing entry surfaced in the search we ran. Verify directly with the provider |
| Vendor regulatory status in Australia | ASIC Connect register search | No matching entry surfaced in the search we ran. Verify directly with the provider |
| Fee schedule for Space and Time Enterprise | Source article | Not disclosed in the source material. Verify with the provider |
| Independent performance or latency benchmark | Source article | Not published in the source material. N/A |
That table is short on purpose. A launch announcement is a press document, not a due diligence file, and treating it as one is how retail traders end up funding accounts on the strength of a headline.
The backtest gap starts with the data, not the model
Here is where we think the announcement is genuinely useful, and where it is oversold.
A verification layer can prove that a calculation over a dataset was performed correctly. It cannot prove that the dataset was the right one, that the strategy that consumed it matches the strategy in the marketing PDF, or that the risk limits described on the website were the limits the code enforced. Those are three separate problems, and only the first one is on the table here.
We have watched this exact pattern in our own review program. When we re-implement a published strategy in our backtest harness and compare it to the vendor's stated rules, the divergences we find are rarely in the math. They are in the assumptions: when the position is sized, which price is used for the fill, what happens on a partial bar, whether the stop is checked intrabar or on close. Proof of SQL does nothing about any of that. It is a tool for verifying computation, not for verifying mandate.
That distinction is the part of this story the enterprise framing tends to bury, and it is the part a retail subscriber should care about most.
What does this change for drawdown control?
Less than you would hope, at least in the short term.
Drawdown behaviour is a function of position sizing, correlation assumptions, and the liquidity of what the bot is trading. A proof layer can tell you that the correlation matrix the bot used was computed honestly from the data it was given. It cannot tell you that the correlation matrix was a sensible input in the first place, and it certainly cannot tell you that the bot will respect a hard equity stop when volatility spikes.
In our funded-account trials, the drawdown events that hurt most are the ones where the bot kept trading through a scheduled event it was not designed for. A verified data feed does not prevent that. A documented kill switch and a hard equity floor do. Those are operational features, not cryptographic ones, and they are the first thing we ask a provider to demonstrate.
This is also where a bot with explicit deviation reporting pulls ahead of the pack. When we compare a reviewed bot against Zephyr AI's strategy-deviation reporting, the difference is not in the proof of computation. It is in whether the subscriber can see, trade by trade, where the live behaviour drifted from the stated specification. That is a governance feature, and it is the one most providers still do not ship.
| Bot failure mode | Does a Proof of SQL-style layer address it? | How named alternatives handle it | What to verify with your provider |
|---|---|---|---|
| Corrupted or mis-sourced market data feed | Partially, if the feed sits inside the verified dataset | Backtrader and NautilusTrader rely on the feed the user supplies, with no built-in provenance layer | Whether historical bars are vendor-sourced or exchange-sourced, and whether they are revised |
| Backtest results that cannot be reproduced | Only if the entire backtest ran inside the verified environment | MetaTrader's Strategy Tester runs locally with no external audit trail | A reproducible backtest export or hash |
| Live behaviour drifting from the stated strategy | No | 3Commas and Cryptohopper publish strategy templates, and their public documentation does not describe a per-trade deviation log | A per-trade deviation report. Verify directly with the provider |
| Broker or API disconnect mid-trade | No | MetaApi and similar bridges focus on reconnection, not verification | The documented reconnect and flatten policy |
| Drawdown exceeding the stated limit | No | RoboForex and comparable brokers publish margin rules, not bot-level drawdown caps | The hard equity stop, in writing |
| Withdrawing funds while positions are open | No | Standard broker withdrawal flow | How open positions are handled at withdrawal |
Free Download: Space and Time Data-Verification Due-Diligence Checklist for AI Trading Agents
A 12-point checklist to confirm whether Space and Time's zero-knowledge proof layer actually verifies the private data, oracle feeds, and model inputs your trading bot acts on before you wire it into live execution.
Get the Verification Checklist
Not sure which AI trading bot fits your strategy? Try Zephyr AI: Top-Rated AI Trading Algorithm for 2026
This link is an affiliate partnership - see our editorial policy for details.
Where the fee model meets the proof layer
Space and Time did not publish pricing for the Enterprise tier in the source material, so we will not guess at it. What we can say is that verification infrastructure has a cost, and that cost has to land somewhere: either on the data vendor, on the bot provider, or on the subscriber.
For a retail trader, that matters because bot subscription economics are already tight. A strategy that clears a few percent a month after fees can be wiped out by an extra monthly line item, and a strategy that clears nothing is not rescued by better data provenance. When we model a bot's net economics, we model the subscription fee, the spread cost, the financing cost, and the slippage as separate line items, because providers routinely quote only the first one. A verification layer that adds a fourth or fifth line item is a real cost, and it should be justified by a measurable reduction in the failure modes above. Right now, we cannot measure that reduction, because nobody has published the before-and-after numbers.
Can you actually switch it off?
This is the question almost nobody asks before subscribing, and the one that matters most when something goes wrong.
An agentic bot typically holds an API key with write access to your account. Disengaging cleanly means more than clicking cancel. It means revoking the key, confirming the bot has no open positions, confirming no resting orders remain, and confirming the provider's system has actually stopped sending instructions. In our review program we treat the disengagement flow as a scored feature, not an afterthought, and we have seen providers where the cancellation page and the trading engine were clearly not talking to each other.
A proof layer does not fix any of this. It is worth saying plainly: the Space and Time launch improves the trustworthiness of the data an agent consumes. It does nothing about the trustworthiness of the agent's authority over your money.
Is any AI trading bot provider actually regulated?
Mostly no, and the register checks take about five minutes.
Space and Time is a data infrastructure company, not a retail broker or an investment firm, so the absence of an entry in the FCA register search or the ASIC Connect register search we ran is expected rather than alarming. But it is a useful reminder that a well-funded vendor with a famous backer is not the same thing as a regulated one.
For the bot providers themselves, the check is straightforward. Search the FCA Register for UK authorisation, the NFA BASIC database for US futures-related registration, SEC EDGAR for filings, the CySEC list of regulated entities for Cyprus, the MAS Financial Institutions Directory for Singapore, and the ESMA registers for the EU passporting picture.
| Regulator | Register to search | What a retail bot user should look for |
|---|---|---|
| FCA (UK) | register.fca.org.uk | Whether the firm is authorised and what permissions it holds |
| ASIC (Australia) | ASIC Connect | AFS licence status and any banning or disqualification records |
| NFA (US) | NFA BASIC | Membership, principal listings, and disciplinary history |
| SEC (US) | SEC EDGAR | Filings, if the provider is a reporting entity |
| CySEC (Cyprus) | CySEC regulated entities | Licence number and permitted activities |
| MAS (Singapore) | MAS Financial Institutions Directory | Whether the provider holds a capital markets services licence |
| ESMA (EU) | ESMA registers | Passporting status across member states |
If a provider is not on any of these and is holding client money or discretionary authority over your account, that is a finding. We treat it as one in every review we publish.
What the proof layer does not prove
Here is the observation we think the launch coverage missed entirely.
Proof of SQL proves that a computation was performed correctly on a dataset. It does not prove that the computation was the one you agreed to. An agent can generate a mathematically flawless proof of a perfectly executed calculation that fed a strategy you never subscribed to, sized a position you never authorised, and held it through a risk event your mandate excluded. Every link in that chain is verifiable. The chain itself is wrong.
In trading terms, this is the difference between a clean audit trail and a clean mandate. The industry has spent a decade building the first and almost no time building the second. When we re-implement a published strategy in our own harness, the divergences we find are almost never arithmetic errors. They are scope errors: the bot traded a symbol the spec excluded, or held overnight when the spec said flat by the close, or sized on equity when the spec said fixed fractional. A proof layer would happily certify every one of those trades.
The providers that will benefit most from verification infrastructure are the ones that also publish a deviation log, because they are the only ones whose subscribers can tell the difference between a verified calculation and a verified strategy.
How Zephyr AI compares
On the specific dimension this launch leaves open, deviation transparency, the comparison is not close. Space and Time is building a way to prove that an agent's arithmetic is honest. Zephyr AI ships a deviation report alongside the live account, which is the mechanism a subscriber actually needs to catch scope drift before it becomes a drawdown.
That is a narrower claim than it sounds, and we want to be precise about it. We are not saying verification infrastructure is unnecessary. We are saying that for a retail account, the governance layer sits closer to the money than the cryptography layer does, and a bot that shows you where it broke its own rules is more useful than a bot that can prove it followed them perfectly into the wrong trade. Where Zephyr AI's adaptive position-sizing has edged out other engines in our review cycle, the advantage has come from the same place: the sizing responds to realised conditions rather than to a static assumption set, and the change is visible in the reporting rather than buried in the engine.
Not sure which AI trading bot fits your strategy? Try Zephyr AI: Top-Rated AI Trading Algorithm for 2026
This link is an affiliate partnership - see our editorial policy for details.
Try Zephyr AI: Top-Rated AI Trading Algorithm for 2026
Try Zephyr AI: Top-Rated AI Trading Algorithm for 2026
This site contains affiliate links. We may earn a commission if you sign up through our links, at no extra cost to you. This does not affect our editorial independence.
Frequently Asked Questions
Does this platform have anything to do with retail trading bots?
Indirectly. Space and Time Enterprise is enterprise data infrastructure, not a trading product. Its relevance to retail is that it is an early attempt to make the data an AI agent consumes independently verifiable, which is one of the inputs to the backtest-to-live gap.
Can an AI trading bot prove its backtest was real?
Not with today's tooling, in most cases. A verification layer can prove that a calculation over a dataset was executed correctly, but it cannot prove the dataset was the right one or that the strategy matched the one advertised. Ask your provider for a reproducible backtest export and verify it yourself.
Does Proof of SQL stop a bot from blowing up my account?
No. Verification addresses data integrity, not position sizing, correlation assumptions, or stop enforcement. The controls that protect an account are a documented hard equity stop, a kill switch, and a broker-side margin regime you understand.
Can I run an AI trading bot on a prop firm account?
Some prop firms permit automated execution and some prohibit it outright, and the rules vary firm by firm. Check the funding partner's terms directly, and confirm whether the firm is registered with a primary regulator before funding anything.
What happens if the API connection drops mid-trade?
That depends entirely on the provider's reconnect and flatten policy, which is rarely published. Ask for it in writing before you subscribe, and confirm whether the bot flattens on disconnect or leaves positions open.
Does this work in the US under Pattern Day Trader rules?
The Space and Time launch is not a brokerage product, so PDT rules do not apply to it. For any bot you run on a US margin account, PDT rules do apply, and a strategy that opens and closes more than three day trades in five business days on an account under the equity threshold will run into them.
How do I check whether a bot provider is regulated?
Search the primary register for the jurisdiction the provider claims. In the UK that is the FCA Register, in Australia it is ASIC Connect, in the US it is NFA BASIC and SEC EDGAR, in Cyprus it is the CySEC list, in Singapore it is the MAS directory, and in the EU it is the ESMA registers.
Can I stop the bot cleanly and withdraw?
Ask three questions before subscribing: how the API key is revoked, whether the provider confirms no open positions remain, and how withdrawals are handled while positions are live. If the provider cannot answer all three in writing, treat that as a finding.
What is the biggest risk the Space and Time launch does not fix?
Scope drift. A bot can produce a flawless proof of a correctly executed calculation that fed a strategy you never agreed to. Verification of computation is not verification of mandate, and only a per-trade deviation log closes that gap.
Not financial advice. Past performance is not indicative of future results. Trading involves substantial risk of loss. Do your own research before making any investment decisions. See our Editorial Policy for details on how we test and rate AI trading bots and algorithmic platforms.
Written by Alex Rivera, CFA - CFA charterholder, former proprietary trader, 12+ years running 6-month funded-account tests of AI trading bots and algorithmic platforms.
Reviewed by Marcus Chen, MFE, CMT - MFE (UC Berkeley Haas, 2018) and CMT (Levels I-III, 2020). Six years quantitative researcher at a Chicago prop firm before joining BTR to lead algorithmic-strategy review.
Read our full Testing Methodology.
More in this category: Trading Industry News.